Managing 100+ Research Interviews: Formats, Naming, and Storage

Qualitative research projects do not fail at analysis; they fail at file management. A dissertation or multi-year study accumulates a hundred-plus interviews recorded across years, devices, and research assistants: M4A from someone's iPhone, WAV from the department's loaner recorder, MP3 from a participant's own recording, the occasional mystery file from a call platform. By year two, nobody is sure which "interview_final_2.m4a" belongs to which participant, half the files will not open in the analysis software, and the corpus that took hundreds of hours to collect is functionally unusable. The cure is a standardization pass — one format, one loudness, one channel layout, one naming scheme — applied ruthlessly at intake.

Why heterogeneous corpora rot

Every inconsistency in a research audio collection compounds over time:

• Format mix breaks tooling. CAQDAS packages, transcription services, and secure repositories each have format quirks; a mixed corpus guarantees that some tool rejects some files at the worst moment.
• Level mismatch poisons review. When one interview whispers and the next blasts, coding sessions become fatiguing and quiet passages get skimmed.
• Ad-hoc names destroy provenance. The link between a file and its participant, consent record, and transcript is the spine of research ethics — and it lives mostly in filenames.
• Uncompressed hoards outgrow storage. A hundred hour-long WAV interviews approach 60 GB, which is when files start getting "temporarily" moved to personal drives and lost.

The standardization pass

One format: MP3

Convert everything at intake. Phone recordings go through an M4A to MP3 converter; recorder WAVs through a WAV to MP3 converter. For speech, 128 kbps is transparent and universally accepted by transcription services and analysis software. If your ethics protocol or archive requires retaining originals, keep them in a sealed "raw" folder — but do all working, coding, and sharing from the standardized MP3s.

One channel layout: mono

Interview audio recorded on a device between two speakers gains nothing from stereo and pays double for it. A stereo to mono pass halves every file and removes the channel-imbalance lottery that phone-on-table recordings produce. The mono vs stereo FAQ covers the reasoning; for a corpus the storage argument alone is decisive.

One loudness

Run every file through a loudness normalizer so the entire corpus plays at one volume. This is not cosmetic: consistent levels make hundreds of hours of coding physically easier, and they measurably improve automated transcription of the quiet interviews.

One size budget

If institutional storage is tight or files must move through a secure portal with upload caps, a compression pass brings outliers into line — the compression FAQ maps bitrate choices to sizes. At 96 kbps mono, a one-hour interview is about 42 MB, and the hundred-interview corpus fits in under 5 GB.

Naming: the scheme that survives

Adopt one pattern before file ten, and never deviate: PROJECT_P###_SESSION_DATE.mp3 — for example "CARE22_P047_S2_2026-03-14.mp3". Participant number rather than name (your ethics board will insist anyway), session number for repeat interviews, ISO date so files sort chronologically by themselves. Keep the key linking participant numbers to identities in the separate, secured location your protocol specifies. Resist the temptation to encode meaning in embedded tags instead: ID3 metadata is useful for display but silently stripped by some platforms, so the filename must carry the identity — see the ID3 FAQ for what tags do and do not reliably preserve. Note also that embedded tags can hold text you did not intend to share; some recorders write device and location details into them, which matters when depositing files in an open archive.

Intake as a ritual

The pass only works if it happens at intake, not as a year-three cleanup. Make it a same-week ritual for every new recording:

• Copy the original into the raw folder (if retained by protocol).
• Convert to MP3, downmix to mono, normalize.
• Trim pre-consent chatter with a trim tool if your protocol requires recordings to begin at consent.
• Rename to the scheme; log the file in the corpus index.
• Back up: working copy, institutional storage, and whatever off-site copy your data-management plan names.

Fifteen minutes per interview, and the corpus never decays.

Pairing audio with transcripts

The standardization pass also fixes the audio-to-transcript link, which is where corpora most often fray. Give every transcript exactly the audio file's name with a different extension — "CARE22_P047_S2_2026-03-14.docx" beside the .mp3 — and the pairing survives any move between machines, coders, and analysis packages without a lookup table. Consistent, normalized mono MP3s also transcribe better and cheaper: services bill by the minute regardless of quality, but clean input means less correction time per hour of tape, and across a hundred interviews the difference is measured in working weeks.

Sharing and deposit

Standardization pays again at the end of the project. Data deposits, supervisor reviews, and inter-coder reliability checks all become trivial when every file is the same format at the same loudness under a predictable name. A collaborator can receive a batch and start coding immediately; an archive can ingest the deposit without a format inventory. The corpus outlives the project — funded studies increasingly require it to — and the difference between a reusable dataset and an unusable one is exactly the pass described here.

A hundred interviews represent something close to a thousand hours of human effort: recruiting, consenting, traveling, listening, transcribing, coding. The standardization pass protects that investment with a browser and a naming convention. It is the least glamorous work in the project, and years later, it is the work you will be most grateful for.